THEORETICAL INVESTIGATION OF THE ROBUSTNESS OF MULTILAYER PERCEPTRONS - ANALYSIS OF THE LINEAR CASE AND EXTENSION TO NONLINEAR NETWORKS

被引：12

作者：

KERLIRZIN, P ^{[1
]}

REFREGIER, P ^{[1
]}

机构：

[1] THOMSON CSF,CENT RECH LAB,OPT GRP,F-91404 ORSAY,FRANCE

来源：

IEEE TRANSACTIONS ON NEURAL NETWORKS | 1995年 / 6卷 / 03期

关键词：

D O I：

10.1109/72.377963

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

In this paper, we address the problem of robustness in multilayer perceptrons. We present the main theoretical results in the case of linear neural networks with one hidden layer in order to go beyond the empirical study we previously made. We show that the robustness can greatly be improved and that even without decreasing performance in normal use. Finally, we show how this behavior, clearly demonstrated in the linear case, is an approximation of the behavior of nonlinear networks.

引用

页码：560 / 571

页数：12

共 15 条

[1] NEURAL NETWORKS AND PRINCIPAL COMPONENT ANALYSIS - LEARNING FROM EXAMPLES WITHOUT LOCAL MINIMA [J].

BALDI, P ;

HORNIK, K .

NEURAL NETWORKS, 1989, 2 (01) :53-58

[2]

BALDI P, 1993, BACK PROPAGATION THE

[3]

COTTRELL DM, 1987, ADV COGNITIVE SCI, V3

[4] THE EIGENVALUES OF RANDOM SYMMETRIC-MATRICES [J].

FUREDI, Z ;

KOMLOS, J .

COMBINATORICA, 1981, 1 (03) :233-241

[5]

GALLINARI P, 1988, JUN INT C NEUR NETW

[6]

Gonzalez R. C., 1977, DIGITAL IMAGE PROCES

[7] NEURAL NETWORKS AND PHYSICAL SYSTEMS WITH EMERGENT COLLECTIVE COMPUTATIONAL ABILITIES [J].

HOPFIELD, JJ .

PROCEEDINGS OF THE NATIONAL ACADEMY OF SCIENCES OF THE UNITED STATES OF AMERICA-BIOLOGICAL SCIENCES, 1982, 79 (08) :2554-2558

[8] ROBUSTNESS IN MULTILAYER PERCEPTRONS [J].

KERLIRZIN, P ;

VALLET, F .

NEURAL COMPUTATION, 1993, 5 (03) :473-482

[9]

KROGH A, 1990, PARALLEL PROCESSING IN NEURAL SYSTEMS AND COMPUTERS, P183

[10]

McClelland J. L., 1986, PARALLEL DISTRIBUTED, V1

← 1 2 →